| Method of extracting a semiconductor device compact model -> Monitor Keywords |
|
|
  Method of extracting a semiconductor device compact modelUSPTO Application #: 20060282802Title: Method of extracting a semiconductor device compact model Abstract: This invention is a method of extracting a semiconductor device compact model by using knowledge of the equations used inside the compact model. Starting by fitting a small subset of the model parameters, the remaining model parameters are fitted and as each new subset of model parameters are fitted, the previously fitted model parameters are adjusted to compensate for the changes introduced due to the currently optimized parameters. This invention details the method of making these adjustments. (end of abstract) Agent: Sitaramarao Srinivas Yechuri - Las Vegas, NV, US Inventor: Sitaramarao Srinivas Yechuri USPTO Applicaton #: 20060282802 - Class: 716004000 (USPTO) Related Patent Categories: Data Processing: Design And Analysis Of Circuit Or Semiconductor Mask, Circuit Design, Testing Or Evaluating The Patent Description & Claims data below is from USPTO Patent Application 20060282802. Brief Patent Description - Full Patent Description - Patent Application Claims
TECHNICAL FIELD OF THE INVENTION [0001] This invention relates generally to electronic testing and characterizing of semiconductor IC processes. More specifically it relates to extraction of compact model parameters from semiconductor device measurement data. BACKGROUND ART [0002] In U.S. Pat. No. 6,779,157 Kondo et al. outline a method use a simulation model to simulate in and calculate local processes that occur during device fabrication and uses the resulting electrical behavior as data to extract BSIM3 model parameters. So the invention is to use the results of simulating a semiconductor device as if it were real data. [0003] In U.S. Pat. No. 6,560,568 Singhal et al. outline a method wherein the results of multiple measurements are itemized in a matrix and the standard deviation is obtained and then the electrical parameters are obtained after normalizing the columns to have a mean of zero and a variance of one. [0004] In U.S. Pat. No. 6,397,172 Gurney et al. outline a method wherein the simulation model is regenerated as required based on the effective simulation conditions and if necessary relevant measurement data relevant to the effective operating conditions are used in regenerating the model. The method also allows selecting a model based on the effective simulation conditions. [0005] In U.S. Pat. No. 6,314,390 Bittner et al. outline a method of obtaining vectors of model parameters from measured data and applying genetic operators and meta evolution operators to create a new population of vectors from which vectors of best fitness are selected. [0006] In a publication entitled "Pre-Silicon Parameter Generation Methodology Using BSIM3 for Circuit Performance-Oriented Device Optimization", IEEE Transactions on Semiconductor Manufacturing, Vol. 14, No. 2, May 2001, Miyama et al. outline a method to extract BSIM3 model parameters before device fabrication. The method is to start with a subset of the BSIM3 parameters considered to be critical and to divide that set into two types namely those which change from generation to generation and those that remain the same. Some of these critical parameters are calculated and some are extracted from data so that finally BSIM3 model parameters are obtained before device fabrication. SUMMARY OF THE INVENTION [0007] There are more than 60 parameters in the BSIM3 model and they cannot all be optimized at the same time. Yet the correct value of many of these parameters depends upon the value of other parameters. So at any point in the optimization sequence you will have a set K of parameters which have been optimized and a set P whose values are yet to be determined. So in this suggested extraction method, parameters from the set K will have to be corrected when parameters from the set P are being optimized. The following is an example. V th = V th .times. .times. 0 - D VT .times. .times. 0 .function. [ e A + 2 .times. e 2 .times. A ] .times. ( V bi - .PHI. s ) ( 1 ) A = - D VT .times. .times. 1 .times. L eff 2 .times. l t ( 2 ) [0008] Consider the case of the threshold voltage V.sub.th using the gate characteristic at a V.sub.ds of 50 mV and zero back bias. When V.sub.th0 is set using the long channel gate characteristic, D.sub.VT0 and D.sub.VT1 are unknown. So a value of V.sub.th0 is set using D.sub.VT0=0. Then when you optimize either D.sub.VT0 or D.sub.VT1, at each iteration you must first increase V.sub.th0 from it's original optimized value by the amount C given by equations 3, 4 where L.sub.effLC is the long channel effective gate lenth, because otherwise the long channel V.sub.th is affected. This correction will not affect the stability of the Levenburg-Marquardt algorithm. C = D VT .times. .times. 0 .function. [ e D + 2 .times. e 2 .times. D ] .times. ( V bi - .PHI. s ) ( 3 ) D = - D VT .times. .times. 1 .times. L effLC 2 .times. l t ( 4 ) BEST MODE FOR CARRYING OUT THE INVENTION [0009] In order to accurately and repeatably extract the BSIM3 model, it is important to compute some of the parameters by solving the model equations using the intermediate variables used inside the model. Obtaining the mobility parameters is done this way as shown below. [0010] In the absence of back-bias all three mobility models in the BSIM3 model have the form of equation 5 where the most often used P is the equation 6. We use three data points from F.sub.jG.sub.0B.sub.0T.sub.n at V.sub.gs=2.times.V.sub.th, V.sub.gs=0.5.times.(V.sub.dd+2.times.V.sub.th) and V.sub.gs=V.sub.dd because the current and electric field are sufficiently low as to avoid the effects of source/drain resistance and mobility degradation. [0011] These 3 points are known to be in the linear region. The effective mobilities .mu..sub.eff.sub.--.sub.1, .mu..sub.eff.sub.--.sub.2 & .mu..sub.eff.sub.--.sub.3 needed at the 3 points points are obtained directly from the model by overriding the intermediate variable .mu..sub.eff inside the model code incrementally until the required I.sub.d is obtained. In this patent application overriding means to substitute your estimated value in place of the value 9 the intermediate variable would otherwise have had. Then you will have the 3 values .mu..sub.eff.sub.--.sub.1, .mu..sub.eff.sub.--.sub.2 & .mu..sub.eff.sub.--.sub.3 and 3 equations of the form of equation 5 at the 3 values P.sub.1, P.sub.2 & P.sub.3, which are solved by matrix inversion to obtain .mu..sub.0, .mu..sub.a and .mu..sub.b. .mu. 0 .mu. eff_n - P n U a - P n 2 U b = 1 ( 5 ) P n = V gsteff + 2 .times. V th T ox ( 6 ) [0012] There are 16 optimization steps in this extraction method. As the parameters are computed in each step, they are set at that value for the following steps. Since the LMA allows a parameter to adjust the significance of each data point, one can use a significance of the drain current at each data point divided by the smallest drain current of the data set. This will allow all data points to be optimized equally. This is done in every LMA iteration. [0013] The first step is to calculate the effective threshold voltage of the 5 MOSFETs from the 15 gate characteristics at minimum V.sub.ds which is obtained from the max slope n and the I.sub.m and V.sub.gs at that point in the gate characteristic as V th = V gs - I m m - V ds 2 ( 7 ) [0014] Set the parameter V.sub.th0 to the V.sub.th obtained for the FET #0 from the curve F.sub.0G.sub.0B.sub.0T.sub.0. [0015] Now we initialize several parameters to their starting values. The parameters set to zero are U.sub.a, U.sub.b, U.sub.c, P.sub.rwg, P.sub.dibl1, P.sub.diblb, C.sub.dsc, C.sub.dscd, C.sub.dscb, N.sub.lx, D.sub.VT0, D.sub.VT1, D.sub.VT2, K.sub.3b, D.sub.VT0w, Eta.sub.0, Eta.sub.b, A.sub.gs, B.sub.0, B.sub.1, N.sub.gate, K.sub.2, A.sub.1, D.sub.VT2w, D.sub.VT0w, C.sub.it. V.sub.off and N.sub.factor. T.sub.oxm is set to the oxide thickness, .mu.=0.1, R.sub.dsw=200, W.sub.r=1, V.sub.sat=150000, P.sub.vag=0.1, P.sub.clm=0.3, P.sub.dibl2=0.2, D.sub.rout=100, P.sub.scbe1=4.24.times.10.sup.10, P.sub.scbe2=1.times.10.sup.-5, K.sub.3=-2.6, W.sub.0=1.times.10.sup.-7, D.sub.VT1w=1, D.sub.sub=1, K.sub.eta=-0.047, Delta=0.01, K.sub.1=0.5, A.sub.2=1 and V.sub.bm=-1.1.times.V.sub.dd. [0016] Now using the drain currents I.sub.j of the j.sup.th FET at the highest V.sub.gs point on the curves F.sub.0G.sub.0B.sub.0T.sub.0, F.sub.1G.sub.0B.sub.0T.sub.0, F.sub.2G.sub.0B.sub.0T.sub.0, F.sub.3G.sub.0B.sub.0T.sub.0, you calculate L.sub.int and W.sub.ift as follows. Increment W.sub.ift from it's minimum expected value to it's maximum expected value in an outer loop in increments of a few nm, and increment L.sub.int from it's minimum expected value to it's maximum expected value in an inner loop in increments of a few nm and compute the error for each j.sup.th FET as compared to the long-wide FET #0 as in eqn 8. err = 1 - I j L effj W effj W eff .times. .times. 0 I 0 L eff .times. .times. 0 ( 8 ) L eff = L drawn - ( 2 L int ) ( 9 ) W eff = W drawn - ( 2 W int ) ( 10 ) [0017] For each combination of W.sub.ift and L.sub.int select the worst err of the FETs #0, #1, #2 and #3, and seek the combination of W.sub.int and L.sub.int that yield the smallest worst err for the FETs. This is now the correct W.sub.int and L.sub.int that is needed. [0018] Now we use the LMA to optimize V.sub.off, C.sub.it, A.sub.0, A.sub.gs, while computing the parameters .mu..sub.0, U.sub.a and U.sub.b from the F.sub.0G.sub.0B.sub.0T.sub.0 characteristic. The data used are the 3 long channel gate characteristics with zero back-bias i.e., F.sub.0G.sub.0B.sub.0T.sub.0, F.sub.0G.sub.1B.sub.0T.sub.0 & F.sub.0G.sub.2B.sub.0T.sub.0. [0019] For each trial set of V.sub.off, C.sub.it, A.sub.0 & A.sub.gs the mobility parameters .mu..sub.0, U.sub.a and U.sub.b are computed prior to computing the error used by the LMA. Now of these 7 parameters that were optimized C.sub.it & A.sub.0 will be updated in subsequent optimization steps but the other 5 parameters are at their final values. Store SC.sub.it=C.sub.it. [0020] Now we use the LMA to optimize U.sub.c, N.sub.factor, K.sub.1, K.sub.2, K.sub.eta, A.sub.0. The parameter C.sub.it is corrected. The data points used are all the gate characteristics for the FET #0 i.e., F.sub.0G.sub.0B.sub.0T.sub.0, F.sub.0G.sub.1B.sub.0T.sub.0, F.sub.0G.sub.2B.sub.0T.sub.0, F.sub.0G.sub.0B.sub.1T.sub.0, F.sub.0G.sub.1B.sub.1T.sub.0, F.sub.0G.sub.2B.sub.1T.sub.0, F.sub.0G.sub.0B.sub.2T.sub.0, F.sub.0G.sub.1B.sub.2T.sub.0 and F.sub.0G.sub.2B.sub.2T.sub.0. C.sub.it is corrected in each iteration of the LMA, prior to computing the error used by the LMA, using the equation 11, where SC.sub.it is the stored value of C.sub.it at the end of Step 4. C it = SC it - N factor .times. q .epsilon. si N ch 2 .PHI. s ( 11 ) Continue reading... Full patent description for Method of extracting a semiconductor device compact model Brief Patent Description - Full Patent Description - Patent Application Claims Click on the above for other options relating to this Method of extracting a semiconductor device compact model patent application. ###  How KEYWORD MONITOR works... a FREE service from FreshPatents1. Sign up (takes 30 seconds). 2. Fill in the keywords to be monitored. 3. Each week you receive an email with patent applications related to your keywords. Start now! - Receive info on patent apps like Method of extracting a semiconductor device compact model or other areas of interest. ### Previous Patent Application: Estimation of average-case activity for digital circuits Next Patent Application: Novel test structure for automatic dynamic negative-bias temperature instability testing Industry Class: Data processing: design and analysis of circuit or semiconductor mask ### FreshPatents.com Support Thank you for viewing the Method of extracting a semiconductor device compact model patent info. IP-related news and info Results in 1.69794 seconds Other interesting Feshpatents.com categories: Daimler Chrysler , DirecTV , Exxonmobil Chemical Company , Goodyear , Intel , Kyocera Wireless , |
||
